DISCUSSION
Ice-cover during the last glacial maximum (LGM; 26.5-19 kya) displaced most high-latitude species, forcing them into ice-free glacial refugia (Hultén 1937; Hewitt 2000, 2003; Bennett & Provan 2008; Clark et al. 2009). The largest documented LGM macro-refugium in North America was located south of the Laurentide and Cordilleran ice sheets in the contiguous U.S., evidenced by fossil data, climatic modeling, and phylogeographic signatures (Graham et al. 1996; Jackson et al. 2000; Holliday et al. 2002), although the extent and complexity of this refugium requires further resolution. Additional LGM refugia are hypothesized in Beringia, the area of exposed continental shelf connecting Alaska to eastern Siberia (Hultén 1937; Abbott & Brochmann 2003; Hope et al. 2013), and multiple smaller micro-refugia are proposed among the archipelagos of the North Pacific Coast (Heaton et al. 1996; Hewitt 2000; Carrara et al. 2003, 2007; Lacourse et al. 2003; Mathewes & Clague 2017), although the duration and possible cyclic recurrence of these refugia remains uncertain. Coastal refugia within the Alexander and Haida Gwaii archipelagos could explain high levels of endemism in this region (Cook & MacDonald 2001; Dawson et al. 2007) and clustered phylogenetic breaks separating insular and continental populations (Colella et al. 2018c; Sawyer et al. 2019) hypothesized to result from post-glacial refugial population expansion limited by secondary contact with closely-related, previously-allopatric taxa (Hewitt 2000). Paleoendemic refugial persistence also explains the rapid reestablishment of complex biotic communities so quickly following deglaciation (Lesnek et al. 2018; Ager 2019).
Significant geographic structure within Pacific martens is consistent with the Coastal Refugium Hypothesis (CRH) (Fig. 3, 4), suggesting the persistence of at least one insular M. caurina population in a North Pacific coastal refugium potentially located along the western fringe of the Alexander or Haida Gwaii archipelagos. The two clades within M. caurina are genetically distinct, parsing an insular and continental lineage, despite mitochondrial nesting (Fig. 2d; Supplemental Information 5). The two insular and continental M. caurina clades are geographically discontinuous (Fig. 1) and estimated to have diverged almost 1 million years ago (Fig. 3b), although PSMC date estimates are highly sensitive to scaling (mutation rate and generation time). Divergence predating the most recent interglacial suggests that insular M. caurina may have diverged from continental populations over multiple glacial cycles, perhaps initially in a coastal refugium and then subsequently on one or more NPC islands. Our genomic results initially contradict the fossil record, which shows a scarcity of fossils on POW Island during the LGM (~20-15 kya, Lesnek et al. 2018) and documents martens appearing on POW during the late Pleistocene (>14 kya) and early Holocene (9-14 kya, Heaton & Grady 2003; Pauli et al. 2015). Absence of martens and other mammals in the Southeast Alaskan fossil record during the LGM may reflect sampling bias, as most dated fossil materials from the region were collected from the Shuká Káa cave at the northern end of POW. Insular M. caurina have not been documented on POW and this site was likely ice-covered at the peak of the LGM (Lesnek et al. 2018). Even so, a number of meso-carnivore teeth from Shuká Káa cave morphologically identified as mink (Mustela vison ) may instead mark the early presence of insularM. caurina (Heaton & Grady 2003), as these species have similar tooth morphology. Similar to misidentifications of Pleistocene coastal black bear (Ursus americanus ) fossils from POW that were originally listed as brown bears (Ursus arctos ) due to size differences over evolutionary timescales (Lindqvist pers. obs.), insular M. caurina are physically larger than bothM. americana and continental M. caurina (Colella et al. 2018b) which may confound taxonomic assignment of dentition. The persistence of diverse communities of large terrestrial mammals, including caribou, bears, and foxes, evident in the fossil record both pre- and post-LGM (Lesnek et al. 2018), points to a higher potential for local refugial persistence through the LGM over the recolonization of these outer islands from mainland sources since the Holocene (Ager 2019).
The viability of a coastal migration route for human colonization of the Americas hinges on our understanding of glacial extent and biotic community composition along the NPC during the late Pleistocene. Geological investigations of southeast Alaska have produced mixed results. Bathymetry (Carrara et al. 2003, 2007) and palynology (Ager 2019) support the persistence of coastal refugia, while cosmogenic exposure dating has shed doubt on hypothesized refugial locations (Lesnek et al. 2018). Multiple signatures of refugial persistence across taxa (Foster 1965; Heaton et al. 1996; Hewitt 2000, 2003; Weckworth et al. 2005; Colella et al. 2018c; Sawyer et al. 2019) is detailing increasingly complex refugial communities along the coast.
For marten, the laterally dilated cranial shape of insular M. caurina hints at a dietary shift towards the consumption of marine prey items (Colella et al. 2018b), also documented in stomach contents (Giannico & Nagorsen 1989) and also reflected in insular wolves of the NPC (Darimont et al. 2009; Muñoz-Fuentes et al. 2010). Martens rely on deep persistent snow and complex forest structure (Proulx 1997; Pauli et al. 2013; Manlick et al. 2017; Martin et al. 2019) for predator avoidance, thermal management, and efficient locomotion, suggesting that refugial ecosystems would have contained forest community assemblages. Access to both marine and terrestrial prey items and timber resources along a NPC migration route, would have enhanced human survivorship during an early pulse of human migration into the Americas via the Pacific coast (Fladmark 1979; Dixon 1993).
The insular-continental biogeographic and phylogenetic break withinM. caurina is largely consistent with signatures from numerous other NPC paleoendemics (bears, Heaton et al. 1996; deer, Latchet al. 2009; ermine, Colella et al. 2018c; shrews, Demboski & Cook 2001; deer mice, Sawyer et al. 2019) and also evident in the few associated parasites examined to date (Soboliphyme baturini , Koehler et al. 2007, 2009; Hoberget al. 2012). Disparate distributions of insular lineages across these heterogeneous archipelagos suggests that the geographic pattern and duration of refugial isolation may vary across climate cycles, depending on the ecological plasticity and dispersal abilities of incumbent species. Insular ‘ABC’ brown bears (Ursus arctos,Heaton et al. 1996), for example, are currently geographically restricted to the three northern most islands of the Alexander Archipelago (Admiralty, Baranof, Chichagof), while the insular black bear lineage (Byun et al. 1997) has a more southerly distribution encompassing southern Alexander Archipelago islands, the Haida Gwaii Archipelago, Vancouver Island, and coastal British Columbia. Under the assumption of niche conservatism, this phylogeographic pattern suggests a cooler, northern refugium within the Alexander Archipelago and a slightly warmer, perhaps more heavily vegetated refugial ecosystem to the south, either in the southern Alexander Archipelago or Haida Gwaii. Early paleoclimatic models for NPC refugia hypothesized these areas to be primarily tundra and unable to support forest-associated taxa such as black bears and martens (Hansen & Engstrom 1996; Ager 2007). More recently however, palynological investigations and radiocarbon dating of postglacial peat and sediment cores indicate that coastal forests similar to today’s forests existed in the Alexander Archipelago during the last interglacial (Ager 2019). Rapid colonization of the western-most islands by pine trees (Picea ) immediately following glacial recession (~17 kya) hints at the potential refugial persistence of coniferous forests (Lesnek et al. 2018; Ager 2019) and parallels our hypothesis that refugial persistence of insularM. caurina is more likely than post-glacial recolonization.
Comparative demography also identified at least three major evolutionary trajectories: M. americana, continental M. caurina (Fig. 3b) and insular M. caurina, consistent with the CRH. Martes americana distributions of effective population size are overall higher than those of M. caurina clades, consistent with the contiguous contemporary range of this species and historical divergence in and subsequent expansion from a single, large eastern refugium (Stoneet al. 2002). Within M. americana , Chichagof Island exhibits the highest effective population size (Ne), followed by central Alaska (Fig. 3b). Although high Neis surprising for an insular population, Chichagof Island received iterative translocations of M. americana in the mid-1900s from multiple source populations, including four other islands in southeast Alaska (Baranof, Wrangell, Mitkof [Petersburg], and Revillagigedo [Ketchikan] islands) and one distant continental locality (Polly Creek, in central AK). These introductions may inflate population size estimates as a consequence of outbreeding (Paul 2009) and make the historical distribution of Ne for this individual resemble that of its source populations (e.g., MAK). In contrast, Prince of Wales Island (POW) received introductions from only two proximate sources: Revillagigedo and Mitkof islands (Burris & McKnight 1973). Revillagigedo Island shows a similar demographic history to POW, suggesting those translocations resulted in successful establishment (Elkins & Nelson 1954). Although our results hint at insular-continental structure within M. americana (Fig. 3), this signal is muddled by historical wildlife translocations and remains unresolved from a nuclear perspective (Fig. 2). Relative to M. americana , both continental and insular M. caurina have persistently smaller effective population sizes.
Among continental species, a common response to rising temperatures is the upward distributional shift in elevation (or latitude) to retain suitable environmental conditions (Hampe & Jump 2011). The fragmented contemporary distribution of continental M. caurina populations (U.S. west coast, Pacific Northwest forests, mountaintops of the American Southwest; Fig. 1) is consistent with the retention of a cooler paleoclimatic niche for species experiencing increasing fragmentation under current warming conditions (Brown 1971; Anderson et al.2000; Parmesan 2006; Hampe & Jump 2011; Meng et al. 2019). Relative to all continental taxa, insular M. caurina show a significantly depressed effective population size through time, with the highest overall inbreeding coefficients. Although likely a consequence of island life, small effective population sizes and high levels of inbreeding place insular martens at an elevated risk of extinction (Frankham 1998; Rybicki & Hanski 2013).
Refugial divergence along the NPC also explains the disjunct contemporary distribution of M. caurina (Fig. 1). Along the NPC,M. caurina inhabits at least three islands; however, Admiralty Island in Southeast Alaska is more than 300 km north of the two insular Canadian populations. Although geographic disjunction across three islands is substantial, the genetic similarly of these island populations points to historical divergence in a single coastal refugium and a potentially more widespread historical distribution of insularM. caurina throughout NPC islands. Higher density sampling across the NPC will be necessary to refine the geographic limits of insular and continental M. caurina clades. Refugial divergence of insularM. caurina is further supported by the persistence of the insular lineage in the Kuiu Island hybrid zone (Dawson et al. 2017), ~20km south of Admiralty Island, relictual signatures ofM. caurina on POW (Pauli et al. 2015), and associated nematodes (Soboliphyme baturini ) on Chichagof Island (Koehleret al. 2007, 2009; Hoberg et al. 2012). Chichagof martens harbor distinctive nematodes that are phylogenetically close to S. baturini found in other populations of M. caurina , suggestingM. caurina or a ‘ghost’ marten lineage may persist or have persisted on this island until relatively recently (Koehler et al. 2007, 2009; Hoberg et al. 2012). Similarly, POW is hypothesized to have been colonized by multiple natural sources (Pauliet al. 2015) and iterative translocations of M. americanato this island were surprisingly successful considering as few as 10 martens (4 females) were introduced to the island (Paul 2009). In contrast, our genomic analyses did not find M. caurina alleles in either of the individuals sequenced from islands that received translocations of M. americana . Instead, we found each island to be genetically aligned with M. americana and their translocation source populations: Chichagof Island with central Alaska and POW with Revillagigedo Island (Fig. 2 and 3). Overall our results suggest thatM. caurina were either not present on these islands prior to translocations or were recently replaced or swamped by introduced or invading M. americana . Interspecific competition, outbreeding, or the introduction of foreign pathogens among other variables may have impacted native M. caurina (Plein et al. 2016; Colellaet al. 2018b; Northover et al. 2018). Ultimately, until additional hybrids are sequenced, our results discourage the translocation of American marten for the genetic rescue or restoration of coastal martens due to potential swamping and emphasize the importance of careful source population selection, as the NPC harbors significant cryptic diversity and complex evolutionary histories.
We detected a hybrid individual collected from each natural mixing zone: Kuiu Island Alaska and western Montana in the northern Rocky Mountains (Table 2; Fig. 2-4; Supplemental Information 9-16). Both hybrids were female, had M. americana mitochondrial haplotypes (Fig. 2d), and mixed nuclear ancestry, with the Montana hybrid containing continentalM. caurina alleles and the Kuiu hybrid containing insularM. caurina alleles (Table 2; Supplemental Information 10-13, 16-17). Both admixed individuals were identified as early generational-stage hybrids (e.g., F1’s or a single generation backcrossed with M. americana , Supplemental Information 11-12) with introgression occurring recently (Supplemental Information 21). Although sample sizes are small, the absence of late-generational hybrids is surprising, especially for the Montana zone which has persisted for many generations (Wright 1953). Detection of only early-generational hybrids is consistent with the presence of hybrid incompatibilities, where F1 hybrids experience a temporary elevation in fitness (heterosis) compared to later generational-stage hybrids (e.g., F2 and beyond) that may suffer outbreeding depression as a result of disrupted co-adapted gene complexes (Todesco et al. 2016). The disruption of co-adapted gene complexes or genes involved in local adaptation via introgression, and particularly loci involved in disease and pathogen resistance (Alibert et al. 1994), may pose a particular challenge to naïve insular taxa. This hypothesis warrants further genomic investigation with fine-scale sampling from within hybrid zones and translocated islands.
Differentiation between insular and continental M. caurina was suggested previously based on reduced-representation genetic approaches (Demboski et al. 1999, 2001; Stone et al. 2002; Smallet al. 2003; Dawson et al. 2017), but the extent of divergence was unknown. A genomic pattern of refugial divergence may be more widespread than previously suspected and additional forest-associated taxa, that are not well represented in the fossil record, may have persisted in NPC refugia. Our results underscore the importance of reevaluating work previously based on one or a few genes, as genomic resolution continues to provide unexpected insight into the evolutionary complexities of coastal refugia (Miller et al. 2012; Colella et al. 2018c) and complex landscapes and holds great promise to unravel complexity across time.